Abstract
This study compares the fit and forecast performance of a selected group of parametric Generalised Autoregressive Conditional Heteroskedasticity GARCH (1, 1) models using various underlying distributions. The GARCH (1, 1) type models are empirically tested on the returns of the All Share Index (ALSI), a diversified portfolio of all the shares on the South African Johannesburg Stock Exchange (JSE). Estimates and forecasts generated by each model are compared and analysed to establish the validity and performance of the models. Forecasts given by the various GARCH (1, 1) models are bootstrapped and the efficiency of the models is also investigated through Value at Risk backtesting. The data used is composed of the returns of the ALSI from the 30th of September 2003 to the 14th of August 2013 and the data frequency is daily data. The best fitting distribution is the skewed normal distribution. With regards to the best fitting GARCH (1, 1) model, the E-GARCH (1, 1) model using the normal distribution performed best. The forecasting analysis showed the outperformance of the E-GARCH (1, 1) model and the best underlying distribution is the student’s t-distribution followed by the skewed normal distribution.
M.Com. (Financial Economics)